• Data centers consume lots of water. The example they start with is that Google's data center in Dalles, Oregon used 355 million gallons of water in 2021, which was 29% of all water consumed in the city (its population is 15K though, but they neglect to mention that).
• By 2023, hyperscale data centers used 66 billion liters of water in the U.S.—triple their volume from less than a decade earlier.
• They quote estimates that ChattGPT consumes 500 mL of water for every 10-50 user prompts (or 10-50 mL per prompt).
• in Ireland datacenters in total draw "over 20% of national electricity", which outstrips the total energy usage of all urban homes in the country.
• In Cerrillos, Chile local residents blocked Google's plans to build a datacenter there after discovering the scale of water use the centers would require (169 liters per second iand its a drought-stricken area).
• Power consumption of data centers is enormous and puts a very high load on energy grids across the world.
My impression is that this is clear example of politically-biased podcast with alarmist and accusatory tone, where none of the facts presented are particularly damning in the grand scheme of things.
In data centers with large water consumption, the water mostly (90%) goes to evaporative cooling (they let hot water turn into vapor carrying away the heat) with the rest going to 10% humidification systems (getting to 40-60% humidity inside to prefect static electricity buildup).
Let's take a moment to recognize what a dream "ecological cost" it is - turning water into vapour - compared to the old timey industries and real environmental problems people have had to deal with. Old timers in Cleveland can tell you how until 1970s (before first serious ecological enforcement), Cuyahoga river running through the city would once in a while BURN WITH FIRE Bible-style from all kinds of unprocessed oil-based waste being dumped by plants and factories on its course. It's the unfortunate reality that many of our vital industrial processes that make our civilization possible rely on dissolving all kinds of most dangereous and toxic compounds in water.
Also, the cost of evaporation cooling is not something fundamental to data centers and is not something that cannot be altered with some known engineering solutions and manageble cost overheads if there is a need. For example, in Belgium they built a two-loop water cooling system that can use industry waste or even sea water. You can also get a fully closed-circuit zero running water cooling system (fridge-style) if you absolutely must.
As for high power consumption and "climate change impact", none of this is specific to data centers. We might like it or not but our society runs on energy and electric power. This whole mindset showcased in this podcast that all energy consumption is something bad and its all about reducing it is so 2010s. I think, its obvious by now that this the road to pure economical if not civilizational suicide. A society that does not prioritize building for plentiful, cheap and hopefully clean energy is doomed to wither and stagnate.
Obviously, to summarise I have to remove the supporting examples, and the dozens of different people being interviewed. To be clear, the journalists aren't personally making all the criticisms, just interviewing other people, so if some of the following seems to contradict itself, that's why.
A fair chunk of the podcast involves explaining the context to a broad audience. You know, explaining what a data centre is, outlining the cloud market and its major players, etc.
The criticisms outlined in the podcast include:
* Data centres produce very few jobs for the communities they're located in.
* They are often built in struggling communities where 'enterprise zones' offer big tax breaks, hoping to attract employers.
* They consume quite a lot of power - not as much as, say, an aluminium smelter, but perhaps as much as 150,000 homes. Few cities have that much spare grid capacity, and some have warned about risks of rolling blackouts.
* 20 percent of Ireland's electricity is used for data centres (they're something of europe's data centre capital due to their attractive tax rates)
* Energy demand at data centres leads to greater emissions at power plants. Even if the data centre contracts to only buy renewable power, that might displace less-eco-friendly buyers of renewables onto non-renewable power sources. And a lot of things like 'carbon credits' are based on rather creative accounting.
* One high-profile data centre (in The Dalles, Oregon) is in a town suffering a drought, and consumes quite a lot of water, considering it's a drought area. Grass on the local golf course is completely dead.
* Land and tax breaks are often acquired through secretive shell companies that insist on secrecy agreements with desperate local governments; in one case the government didn't even know they were dealing with Google. This secrecy extends to agreements about things like water usage.
* As you can imagine, a local community suffering a drought sees the local data centre's water consumption being kept secret by elected officials, they assume the worst.
* Some data centre builders, like Elon Musk, have a history of making legally non-binding promises, then not bothering to keep them. And of running large gas generators without permits.
* The kind of distressed post-industrial communities that welcome data centres often have high levels of pollution and cancer, making those unpermitted generators particularly bad.
* Many of the hyperscalers are also big AI boosters, so it's not like the datacentre operators can disclaim responsibility for the power needs of AI.
* Many people have criticisms of AI, beyond energy consumption. Such as huge centralised LLMs transferring more control to huge tech firms; getting things wrong; AI friends being an alienating concept; having heavy-handed censorship; widespread use of bots on platforms like twitter and reddit; risks of job losses; being trained on pirated ebooks without authors' permission; being a really shitty therapist; producing mediocre art; producing porn depicting real people without their consent; producing creepy underage porn.
* Or AI might be a bubble that's about to burst, which would also be bad but for other reasons.
* Tech business leaders like Sam Altman are on record saying some pretty wacky things about AI power consumption, like that the high power demands of AI will force us to invent fusion power. A load of them also have weird, messianic ideas about "the singularity", or think we're all in a simulation already, or think living in a Matrix-style simulated 'metaverse' sounds like a great thing.
* Many of the highest-profile tech folks - the billionaires - have very right-wing politics. Such as opposing all regulation as a matter of principle, except on the occasions when it works for their benefit. Some people think expecting these folks to regulate themselves isn't the best idea.
Overall this is all stuff that followers of tech industry news will probably have heard before; the podcast just adds context, draws it together, and finds sources in the form of interviewees.
Uplifting everyone ensures that we'll be that much more likely to find the next Mozart or Tesla or Torvalds or whoever, if we give them a chance.
But yes, better to acknowledge how capital can be better utilized. You can probably give away free school lunches for an entire generation of children with that $10 billion in Louisiana, or you can give it to Zuckerberg to get slightly richer.
Becomes abundantly clear which one is better for societal advancement.
From another perspective, compared to a coal mine or a paint factory or a steelworks or an airport or a landfill or an oil refinery, data centres are safe, low-pollution, low-noise, low-odour, and low-traffic. By the standards of industrial areas, they're great neighbours.